Reversing Pathologically Increased EEG Power byAcoustic Coordinated Reset Neuromodulation

Ilya Adamchic,1* Timea Toth,1 Christian Hauptmann,1 andPeter Alexander Tass1,2

1Institute of Neuroscience and Medicine—Neuromodulation (INM-7), J€ulich Research Center,J€ulich, Germany

2Department of Neuromodulation, University of Cologne, Cologne, Germany

r r

Abstract: Acoustic Coordinated Reset (CR) neuromodulation is a patterned stimulation with tonesadjusted to the patient’s dominant tinnitus frequency, which aims at desynchronizing pathologicalneuronal synchronization. In a recent proof-of-concept study, CR therapy, delivered 4–6 h=day morethan 12 weeks, induced a significant clinical improvement along with a significant long-lastingdecrease of pathological oscillatory power in the low frequency as well as c band and an increase ofthe a power in a network of tinnitus-related brain areas. As yet, it remains unclear whether CR shiftsthe brain activity toward physiological levels or whether it induces clinically beneficial, but nonethelessabnormal electroencephalographic (EEG) patterns, for example excessively decreased d and=or c. Here,we compared the patients’ spontaneous EEG data at baseline as well as after 12 weeks of CR therapywith the spontaneous EEG of healthy controls by means of Brain Electrical Source Analysis sourcemontage and standardized low-resolution brain electromagnetic tomography techniques. The relation-ship between changes in EEG power and clinical scores was investigated using a partial least squaresapproach. In this way, we show that acoustic CR neuromodulation leads to a normalization of theoscillatory power in the tinnitus-related network of brain areas, most prominently in temporal regions. Apositive association was found between the changes in tinnitus severity and the normalization of d and cpower in the temporal, parietal, and cingulate cortical regions. Our findings demonstrate a widespreadCR-induced normalization of EEG power, significantly associated with a reduction of tinnitus severity.Hum Brain Mapp 35:2099–2118, 2014. VC 2013 The Authors Human Brain Mapping published by Wiley Periodicals, Inc.

Key words: tinnitus treatment; desynchronization; electroencephalography; non-invasiveneuromodulation; phantom perception

r r

INTRODUCTION

Subjective tinnitus is a frequent sensation of sound thatcannot be attributed to an external sound source. Tinnitusmay sufficiently affect everyday life, leading to sleep disor-

ders, work disability, and psychiatric problems [Gopinath

et al., 2010; Langguth et al., 2011; Schutte et al., 2009]. It is

generally accepted that tinnitus generation has a central ba-

sis, typically being initiated by damage to the peripheral

hearing system [De Ridder et al., 2011a,2011b; Eggermont

and Roberts, 2004; Weisz et al., 2005, 2006].

This is an open access article under the terms of the CreativeCommons Attribution License, which permits use, distributionand reproduction in any medium, provided the original work isproperly cited.Additional Supporting Information may be found in the onlineversion of this article.

*Correspondence to: Ilya Adamchic, Institute of Neuroscience andMedicine–Neuromodulation, J€ulich Research Center, Leo-Brand-Straße, J€ulich, Germany. E-mail: i.adamchic@fz-juelich.de

Received for publication 12 November 2012; Revised 24 February2013; Accepted 8 April 2013.

DOI: 10.1002/hbm.22314Published online 1 August 2013 in Wiley Online Library(wileyonlinelibrary.com).

r Human Brain Mapping 35:2099–2118 (2014) r

VC 2013 The Authors Human Brain Mapping published by Wiley Periodicals, Inc.

In most cases, tinnitus is associated with an audiometri-cally measurable hearing loss that largely coincides with thetinnitus frequency range [Norena et al., 2002]. However,also in patients with a normal audiogram the presence oftinnitus may be accompanied by abnormal inner hair-cellfunction [Weisz et al., 2006]. Both human and animal datashow that deafferentation alters receptive fields [Dietrichet al., 2001; Irvine et al., 2001; Rauschecker, 2005] andleads to the emergence of pathological neural synchrony[Hauptmann and Tass, 2007; Merzenich et al., 1984; Ochiand Eggermont, 1997; Verkindt et al., 1995; Weisz et al.,2005] in brain regions deprived of peripheral input.Indeed, pathologically enhanced neuronal synchronizationwas observed in the primary auditory cortex of animalsfollowing damage to the inner ear [Hauptmann and Tass,2007; Merzenich et al., 1984; Ochi and Eggermont, 1997]as well as in tinnitus patients [De Ridder et al., 2011b; Lli-nas et al., 1999; Weisz et al., 2005, 2007b].

In a magnetoencephalography (MEG) study, Weisz et al.[2005] showed a reduction of the a rhythm and a concomi-tant increase of slow-wave (d) activity, particularly in tem-poral regions, in a group of individuals with chronictinnitus compared to tinnitus free controls. In fact, low-fre-quency oscillations are typical for cortical regions deprivedof afferent input [Steriade, 2006]. However, low frequen-cies are not, in general, pathological as they regularlyoccur during slow-wave sleep [Benoit et al., 2000; Steriade,2006]. In contrast, in patients suffering from chronic sub-jective tinnitus, one observes persistent low-frequency ac-tivity, present in the awake state. Further studies revealedthat positive symptoms may arise owing to the concertedaction of slow- and high-frequency oscillations [De Ridderet al., 2007; Llinas et al., 1999; Weisz et al., 2007b]. Slow-wave activity facilitates and sustains c activity that, inturn, may serve as a neural code of auditory phantom per-ception [De Ridder et al., 2007; Weisz et al., 2007b]. Clini-cal significance of this pattern of electroencephalogram(EEG) abnormality is confirmed by the fact that the tinni-tus-related distress and tinnitus loudness are correlatedwith this abnormal spontaneous activity pattern [De Rid-der et al., 2011b; van der Loo et al., 2009; Weisz et al.,2005]. Apart from auditory cortical brain areas, nonaudi-tory areas involved in attention and emotional regulationwere also shown to be involved in the tinnitus generation,in particular, in patients with considerable amount of tin-nitus distress [De Ridder et al., 2011a; Vanneste et al.,2010]. Furthermore, an altered functional connectivitybetween auditory and nonauditory regions seems to be ahallmark of the auditory phantom perception [De Ridderet al., 2011a; Schlee et al., 2009a]. However, these pioneer-ing reports of altered EEG=MEG rhythmicity in tinnituswere related to a comparison between a group of tinnituspatients and a group of tinnitus free controls [Moazami-Goudarzi et al., 2010; Weisz et al., 2005]. One limitation ofthese studies is that, contrary to the normal hearing con-trol group, tinnitus patients typically have an audiometri-cally measurable hearing loss [Weisz et al., 2005].

Consequently, it was unclear whether the observed EEGabnormalities were specific to tinnitus, rather than beinggenerated by the sensory deprivation owing to the hearingimpairment [Weisz et al., 2005]. This point is particularlyrelevant as the findings by Weisz et al. (2005) are, to a cer-tain extent, qualitatively similar to EEG findings obtainedduring slow-wave sleep [Mikhailov, 1990]. Accordingly,further attempts to study the electrophysiological correlateof tinnitus in humans focused on intervention-relatedchanges of brain oscillations within one patient group (forreview see, Langguth et al., 2012), for example, by meas-uring neurophysiological effects of tinnitus maskers [Kahl-brock and Weisz, 2008], auditory cortex stimulation viaimplanted electrodes [De Ridder et al., 2011b] or neuro-feedback training [Dohrmann et al., 2007b]. In an MEGstudy, Kahlbrock and Weisz (2008) found a significantreduction of d-band activity in temporal areas during re-sidual inhibition following the offset of tinnitus maskerapplication [Kahlbrock and Weisz, 2008]. However, no tin-nitus-free controls were used in these interventional stud-ies. It is not clear whether intervention-induced tinnitusrelief is necessarily related to a normalization of the EEGpattern. In principle, brain oscillations might be modifiedin such a way that tinnitus decreases, whereas EEG pat-terns are changed, but still remain significantly differentcompared to healthy controls.

In a previous study, we analyzed clinical and EEG changescaused by acoustic Coordinated Reset (CR) neuromodulationwithin a population of patients with chronic subjective tinni-tus [Tass et al., 2012a]. As shown computationally, CR neuro-modulation specifically counteracts pathological neuronalsynchronization by desynchronization [Tass, 2003]. Changesof neuronal dynamics and synaptic connectivity are stronglylinked in dependence on the relative timing of the pre- andpostsynaptic spikes by the spike timing-dependent plasticity(STDP) [Gerstner et al., 1996; Markram et al., 1997]. Alreadyin simple neuronal networks comprising STDP, stronglysynchronized states with strong synaptic connectivity maystably coexist with desynchronized states with weak synapticconnectivity [Tass and Hauptmann, 2009; Tass and Majtanik,2006]. CR-induced desynchronization decreases the rate ofcoincidences and, hence, owing to STDP also the averagestrength of the synaptic connections [Tass and Hauptmann,2009; Tass and Majtanik, 2006]. Consequently, the stimulatedneuronal population is shifted from a synchronized statewith strong synaptic connectivity to a desynchronizedstate with weak connectivity [Tass and Hauptmann, 2009;Tass and Majtanik, 2006]: The network undergoes an anti-kindling, that is, it unlearns pathological connectivity andpathological synchrony. As shown in a computationalstudy, according to the underlying biophysics, thelong-lasting desynchronization and the unlearning ofpathological connectivity (antikindling) can robustly beachieved by means of direct electrical CR stimulation or indi-rect, that is, synaptically mediated, excitatory, andinhibitory CR stimulation [Popovych and Tass, 2012].

r Adamchic et al. r

r 2100 r

Electrical CR neuromodulation caused long-lasting desynch-ronization in rat hippocampal slice rendered epileptic by mag-nesium withdrawal [Tass et al., 2009] and sustained long-lasting therapeutic aftereffects in MPTP monkeys [Tass et a.,2012], the standard model of experimental parkionsonism.

To specifically counteract pathological neuronal syn-chrony in the central auditory system, we used acousticCR neuromodulation [Tass et al., 2012a]. To this end,based on the tonotopic organization of the central auditorysystem, sequences of pure tones with pitches centeredaround the patient’s dominant tinnitus frequency are peri-odically delivered in an ON–OFF protocol (METHODS)[Tass and Popovych, 2012; Tass et al., 2012a]. Treatmentwith CR neuromodulation resulted in a highly significantdecrease of tinnitus symptoms as measured by visual ana-log scale (VAS) and tinnitus questionnaire (TQ) scores[Tass et al., 2012a]. According to VAS [Adamchic et al.,2012a] and TQ [Adamchic et al., 2012b] evaluation studies,this improvement is not only statistically, but also clini-cally significant. In contrast, placebo treatment did notlead to any significant changes. Furthermore, after 12weeks of acoustic CR neuromodulation d and c activitywere significantly decreased in primary and secondary au-ditory cortex and in frontal areas combined with anincrease of the initially reduced a power in auditory andprefrontal areas [Tass et al., 2012a]. However, in thatstudy, EEG markers were compared in one patient popula-tion before and after CR therapy [Tass et al., 2012a].Strictly speaking, it was, thus, not possible to judgewhether the therapy shifted the EEG markers closer towhat is supposed to be physiological (as represented by acontrol group of healthy subjects) or whether a completelydifferent pattern of EEG markers evolved, for examplecharacterized by significantly reduced d and=or c power asopposed to controls.

Although neural synchronization plays a fundamentalrole in the pathophysiology of the auditory phantom per-ception, up to our knowledge, as yet no study has investi-gated therapy-induced changes of EEG patterns in tinnituspatients as compared to physiological reference EEG pat-terns recorded from a group of tinnitus-free controls. In thisstudy, we set out to overcome this shortcoming. To furtherour understanding of the pathophysiology of chronic sub-jective tinnitus and of the mechanisms of acoustic CR neuro-modulation, the goals of this study are as follows: (i) Tostatistically discriminate between na€ıve tinnitus patients

and tinnitus-free controls on the basis of EEG spectral pa-

rameters. (ii) To compare the EEG pattern in the tinnitus

patient population after 12 weeks of CR therapy with the

EEG pattern of the tinnitus-free controls. (iii) To explore

relationships between CR therapy-induced changes of dif-

ferent resting EEG parameters (i.e., power changes in differ-

ent frequency bands observed in different brain areas) on

the one hand and tinnitus symptoms on the other hand. In

particular, to study whether acoustic CR neuromodulation

normalizes the EEG pattern in the tinnitus patients (i.e.,

shifts the EEG patterns closer to a physiological pattern), or

whether clinical improvement is associated with a signifi-

cantly different, nonphysiological EEG pattern.To our knowledge, this is the first study that investi-

gates whether treatment-induced long-lasting changes ofresting EEG spectral parameters contribute to a normaliza-tion of oscillatory brain activity and if so in which corticalbrain regions normalization of EEG spectral parameterstakes place. To test this, we combined a cross-sectionalwith a longitudinal approach. To assess the relationshipbetween the normalization of oscillatory activity in differ-ent brain areas and the reduction of tinnitus severity, weused the partial least-squares (PLS) multivariate approachthat holds specific advantages over conventional univari-ate approaches [McIntosh et al., 1996].

MATERIALS AND METHODS

Patients

Here, we analyze EEG data recorded in tinnitus patientswho participated in a multicentric randomized, controlledclinical trial on Acoustic CR Neuromodulation in theTreatment of chronic subjective tonal tinnitus, performedin Germany between 2009 and 2010 (“RESET study,” Clini-calTrials.gov Identifier: NCT00927121). In total, 63 patientswith chronic subjective tonal tinnitus participated in theRESET study. All patients were informed about the scopeand aim of the study and a written consent was obtainedfrom all tinnitus patients according to the declaration ofHelsinki and the study was approved by the ethics com-mission. Patients with pulsatile, ringing, buzzing, roaring,or hissing tinnitus as well as patients with a need for hear-ing aids, a history of auditory hallucinations, M�eniere’sdisease, diagnosed neurological or mental disorders, andpatients taking CNS-acting medication or participating inother tinnitus therapy programs were not included in thestudy [Tass et al., 2012a]. The extent of the hearing losswas investigated with a pure tone audiogram. Patientswith a hearing loss at any of the tested frequencies (i.e.,0.125, 0.250, 0.750, 1, 2, 3, 4, 6, 8, 12 and kHz) >50 dBwere not included in the study. By the same token,patients who were not able to hear all CR therapy tones(see below) were excluded from the study. In brief, 1–4and 6–12 kHz pure-tone averages (PTAs) were calculated.

From 63 randomized patients, 61 had EEGs recorded atbaseline and 12 weeks. In all, 11 out of these 61 patientswere excluded from the analysis as their EEG recordingswere performed with lower hardware filter settings (0.1–50Hz), not allowing for an analysis of higher c-band activity.Unilateral and bilateral tinnitus patients can have differentEEG abnormalities [Vanneste et al., 2011a]. Accordingly, toavoid an influence of such differences, we selected onlythe patients with bilateral tinnitus (n 5 28). An overviewof the patient’s baseline characteristics is summarized inTable I.

r Reversing Pathological Neural Activity in Tinnitus r

r 2101 r

CR Treatment

In the RESET study, patients were stimulated for 12weeks using a portable acoustic device and comfortable ear-phones (for a more detailed description of the CR neuromo-dulaton treatment, see Tass et al., 2012a). Visits took placeafter 1, 4, 8, 12, and 16 weeks. Data for this article wereobtained from the baseline and the 12-week visit, becauseEEG recordings for all patients were performed off-stimula-tion (i.e., at least 2.5 h after cessation of CR neuromodula-tion) at these visits. Subjectively perceived tinnitus loudnessand tinnitus annoyance were assessed off-stimulation usinga visual analog scale for loudness (VAS-L) and annoyance(VAS-A). In general, the CR treatment resulted in a highlysignificant and clinically relevant decrease of tinnitusseverity as measured by VAS-L=VAS-A and TQ scores[Adamchic et al., 2012a,2012b; Tass et al., 2012a].

Healthy Controls

The control group consisted of 16 healthy tinnitus-freesubjects (10 men and 6 women) age matched (mean age45.0, SD 12.5; P 5 0.29) to the group of 28 tinnitus patientsselected for the EEG analysis. Recruitment of study partici-pants, included in the control group, was performed byadvertisement and all participants signed informed con-sent form. Participants, selected to the control group, werescreened by physicians for neurological and mental disor-ders as well as for ear disorders. Subjects taking CNS-act-ing medication were excluded.

The PTAs of 1–4 and 6–12 kHz were 12.73 (13.69) and24.29 (19.05), respectively. All healthy subjects were nottaking any medication known to affect the EEG. Thisgroup was selected to confirm tinnitus-related EEG abnor-malities reported previously [De Ridder et al., 2011b; Moa-zami-Goudarzi et al., 2010; Weisz et al., 2005].Furthermore, the control group served as a reference forthe treatment-induced EEG changes.

Patient Groups for EEG Analysis

We performed three different comparisons: (i) We com-pared all 28 patients with bilateral tinnitus before CR ther-apy with the healthy control group. (ii) We compared all28 patients with bilateral tinnitus after 12 weeks of CRtherapy with the healthy control group. (iii) To investigatesymptom-related changes in the oscillatory brain activity,we also investigated a subgroup of patients with a goodclinical response defined as TQ improvement > 12 points(n 5 12) as they were expected to display the most pro-nounced EEG changes. A TQ > 12 cut-off was selectedbased on the Reliable Change Index [Jacobson and Truax,1991] with the assumption that it separates patients withmoderate to good relief of their tinnitus symptoms frompatients with small to no relief in symptoms [Tass et al.,2012a; Turner et al., 2010].

The group of all 28 patients with bilateral tinnitus com-prised patients from all therapy Groups G1–G4: 10 fromG1, 4 from G2, 6 from G3, and 8 from G4. The group ofgood responders (n 5 12) consisted of six patients fromG1, four from G3, and two from G4.

EEG Data Acquisition

Patients

Every patient underwent two recording sessions: first onthe first treatment day before start of the treatment; secondat the 12-week visit, minimum 2 h after stopping the laststimulation session.

Healthy controls

In every healthy control subject, one EEG recording wasperformed. Patients and healthy controls were instructedto retain from caffeinated beverages on the day of the re-cording. Patients and controls were seated in an uprightposition in a comfortable chair. EEG recordings wereobtained in a dimly lit room in a Faraday cage. EEG datawere collected from 128 surface electrodes (128 channelHydroCel Geodesic Sensor Net). All electrodes were refer-enced to Cz. The EEG signals were amplified with a NetAmps 200 amplifier (Electrical Geodesis, Eugene, OR),digitized at 1 kHz and band-pass filtered from 0.1 to 400Hz. Recordings were performed during awake state witheyes closed and eyes open alternating 2-min long condi-tions. We selected the eyes closed data (two 2-min longepochs) for further analysis as they were less affected byartifacts. Subjects were video monitored for behavioralsigns of drowsiness and the epochs with signs of drowsi-ness were excluded from further analysis. We were inter-ested in spontaneous activity; therefore, we excluded thetransitions between eyes-open and eyes-closed phases byremoving the first and the last 5 s of each eyes-closedepochs. Photogrammetry was performed for all subjectsusing the Geodesis Photogrammetry system (Electrical

TABLE I. Baseline characteristics of all patients and the

patients with a good clinical response (i.e., TQ improve-

ment � 12 points)

All patients(n 5 28)

Patients witha good clinical

response (n 5 12)

Age (years) (SD) 50.0 (10.5) 49.3 (8.9)Tinnitus duration

(years) (SD)6.1 (4.7) 7.8 (5.6)

TQ (SD) 45.6 (16.8) 52.3 (17.5)VAS-L (SD) 68.4 (19.2) 72.1 (20.7)VAS-A (SD) 67.7 (20.5) 71.7 (23.1)1–4 kHz pure-tone

average14.83 (9.99) 14.42 (10.32)

6–12 kHz pure-toneaverage

36.55 (17.65) 33.19 (13.29)

r Adamchic et al. r

r 2102 r

Geodesis, Eugene, OR) and the individual head shape wasmodeled for each subject and EEG session.

Data Analysis

The scalp EEG was rereferenced to average reference.Signals were additionally digitally filtered with a 0.8–130Hz digital filter. Each EEG recording was corrected forblink and eye movements in Brain Electrical Source Analy-sis (BESA) using the surrogate model approach in BESA(MEGIS Software, 5.2 version) [Ille et al., 2002]. Recordingswere further analyzed in MATLAB (The Mathworks,Natick, MA) using EEGLAB (http:==sccn.ucsd.edu=eeglab)where artifact rejection was performed. First, epochs con-taining large non-neural artifacts (e.g., large EMG artifacts,movement artifacts) were removed in EEGLAB. Then, in-dependent component analysis decomposition was per-formed on the sensor level EEG data [Delorme et al.,2007]. Myogenic components, that is, components contain-ing EMG activity in the absence of any identifiable neuro-genic activity, were selected. These components wereidentified based on the following criteria: (i) high broadpeaks around either 30–40 Hz and higher, (ii) a moder-ately small and clustered distribution on the topographicmaps that mimicked the underlying scalp musculature,(iii) periods of high-frequency activation in the time do-main, and (iv) equivalent dipole(s) located outside thebrain volume and having a residual variance of 15% orless. Dipole locations were modeled using DIPFIT plug-inin EEGLAB. Head shape and electrode locations weremodeled for each subject and EEG session separately usingthe EGI Photogrammetry system and then imported intothe DIPFIT plug-in. Myogenic component selection proce-dure was performed twice with inter-rater reliabilityassessed using Krippendorff’s alpha (a 5 0.97). The meanlength of the recordings after artifact correction was 3 min36 s 6 24 s. Surface EEG was transformed into brainsource activity using the source montage approach inBESA [Scherg et al., 2002]. A source model was generatedwith regional neural sources placed in the regions of inter-est (ROIs) using BESA. The source montage consisted oftemporal (T), orbitofrontal (OF), dorsolateral prefrontalcortex (DPFC), and parietal (PA) sources in both hemi-spheres, one source was also located within the anteriorcingulate cortex (CA) and one in the posterior cingulatedcortex (CP). We have to point out that the source montageapproach used here should not be confused with a finespatial localization of, for example ERP, activity performedin some functional neuroimaging studies for each patientand condition. Location of regional sources (ROIs) waspredefined by the authors based on the results of the pre-vious studies and was the same for every patient and re-cording [Lanting et al., 2009; Schlee et al., 2009a; Vannesteet al., 2011b; Weisz et al., 2005]. Additional probe-sourceswere placed into the occipital lobe and in the area of thecentral sulcus in both hemispheres. Sources outside the

ROIs acted as a spatial filter and reduced the contributionof these regions to the ROI. The strength of the sourcemontages approach is that one can obtain time courses ofbrain activities from distinct brain regions [Scherg et al.,2002]. For the PLS statistical analysis, the temporal sourcewas further subdivided into the two sources: (1) an equiv-alent dipole modeling the primary auditory cortex (ROI:AC1) with Thalairach coordinates [x,y,z; mm] 640, 226,212 and orientations: 60.3, 20.5, 20.8 left and right [Ver-kindt et al., 1995]; (2) the secondary auditory cortex (AC2)was modeled with a dipole having a radial orientation[Hegerl et al., 1994]. Normalized band powers from 5ROIs (AC1, AC2, OF, DPFC, and PA) within the left andright hemispheres were averaged over hemispheres foreach ROI and every patient separately, resulting in sevenROIs for 28 patients [Kahlbrock and Weisz, 2008]. Thisaveraging was justified by standardized low-resolutionbrain electromagnetic tomography (sLORETA) statisticalmaps that showed a similar spatial distribution of statisti-cally significant differences in both hemispheres. Further-more, in tinnitus patients altered spontaneous activitypattern in the temporal region was found to be bilaterallysymmetrical, and the same holds true also for activationduring the acoustic stimulation paradigm [Smits et al.,2007; Weisz et al., 2005]. Thus, no unilateral effects wereaffected and=or masked through averaging over bothhemispheres. The fast Fourier transform was performedon the artifact-free source waveforms after windowing thesignal with a 4,096 ms wide cosine squared (cos2) windowwith 50% overlap. This gave us a frequency resolution of0.244 Hz.

The following frequency bands were defined: 1–3.5 Hz(d), 4–7.5 Hz (h), 8–12 Hz (a), 12.5–30 Hz (b), 30.5–48 Hz(low c), and 52–90 Hz (high c). In all calculations, weexcluded the power line artifact (48–52 Hz). Taking intoaccount the comparatively poor test–retest reliability forabsolute as compared to relative power bands, we per-formed our analysis based on the relative power features[John et al., 1983]. The individual power spectra were nor-malized by dividing power at each frequency by the inte-gral of the power across all frequencies from 1 to 90 Hz.This allowed us to compare subjects with large differencesin the overall spectral energy and to estimate the relativecontribution of each band to the whole spectrum and tocompare the relative contribution of different bands. Forthe multivariate analysis, power spectra derived withBESA source montage analysis were divided into 1-Hzwide bands in the range from 1 to 90 Hz.

As shown in Figures 5 and 6, for a better presentation,these 1-Hz wide bands were labeled by abbreviations con-sisting of the lower edge of the frequency band followedby the ROI. For instance, 25DPFC stands for the frequencyband between 25 and 26 Hz in the DPFC cortex. We usedsLORETA to confirm the results received with BESAsource montage [Pascual-Marqui, 2002]. With sLORETA,we computed a three-dimensional linear inverse solutionto the EEG inverse problem with a three-shell spherical

r Reversing Pathological Neural Activity in Tinnitus r

r 2103 r

head model registered to the Talairach human brain atlasdigitized at the Brain Imaging Center of the Montreal Neu-rological Institute [Pascual-Marqui, 2002]. The solutionspace was constrained to the gray matter voxels thatbelonged to cortical and hippocampal regions (a total of6,430 voxels at a 5-mm spatial resolution). The localizationof the differences in current density power between thegroups was assessed by voxel-by-voxel t-tests of the sLOR-ETA images. In the resulting statistical three-dimensionalmaps, a nonparametric approach was used to identify stat-istically significant differences of cortical voxels [Nicholsand Holmes, 2001]. Briefly, if there were no significant dif-ferences between groups, any labeling of voxels wouldresult in an equally likely statistical map. Thus, sLORETAstatistical maps were randomly relabeled and t-valueswere recalculated. Under the null hypothesis, each of thet-statistics are equally likely, and thus, the resulting P-value is the proportion of the t-statistic values greater thanor equal to the t-statistic of the correctly labeled data (formore details, see Nichols and Holmes, 2001). P-valueswere derived from 5,000 such permutations. The voxels’ P-values of 0.05 were colored in a MRI template.

Statistical Analysis of Spectra

For comparisons of power spectra between groups, anonparametric Wilcoxon rank sum test for each frequencypoint was used. This statistical test for frequency pointcomparisons was corrected for the number of tests con-ducted using the false discovery rate (FDR, Benjamini andHochberg, 1995).

Multivariate Analysis

To determine the relationship between power spectrachanges (calculated from BESA ROIs) and changes in clini-cal scores, we applied the PLS analysis [Krishnan et al.,2011]. Power changes in 1-Hz-wide frequency bands wereused for multivariate analysis as predictor variables ofchanges in TQ scores and VAS values. PLS analysis wasselected as it overcomes the problems related to multicolli-nearity and having too many variables as compared to thenumber of samples.

All 28 patients were included into the PLS model. Thepower spectra were divided into 1-Hz wide bands startingat 1 to 90 Hz, which yielded a total of 85 1-Hz wide bands.The power in each band was normalized to the range of1–90 Hz total power. Changes from the first session to thesecond session were calculated for each 1-Hz band and ev-ery patient separately. The power values for the multivari-ate analysis were structured in an X-matrix with one rowper subject (i.e., a total of 28 rows) and one column per 1-Hz-wide frequency band for each source (i.e., a total of595 columns).

The VAS and TQ data were structured into the Y-matrixin an analogous way. For this analysis, we have used

VAS-L and VAS-A values as well as TQ total, psychologi-cal distress (PD), intrusiveness (I), auditory perceptionaldifficulties (A), and sleep disturbance (Si). We did notinclude emotional distress (E) and cognitive distress (C)subscales as they are already represented in PD (PD 5 E1 C) and somatic complaints (So) as it is difficult to ruleout the real origin of somatic complaints (i.e., the Y-matrixconsisted of 28 rows and 7 columns). Several dependentvariables can be modeled at the same time using PLSallowing for a simpler interpretation of the results. How-ever, if the dependent variables are reasonably independ-ent, computing a single PLS model tends to have manycomponents and to be difficult to interpret. In such a case,a separate modeling of clusters of correlated dependentvariables (Ys) results in a set of simpler models with fewerdimensions. Hence, in the process of creating a PLS modelone should start with a principal component analysis(PCA) of the matrix of dependent variables (i.e., Y-matrix)to characterize patterns or “structure” in data [Wold et al.,2004]. If dependent variables cluster in groups in the PCAloading plot, separate PLS models should be consideredfor each cluster [Wold et al., 2004]. Accordingly, first weperformed a PCA on the Y-matrix [Jolliffe, 2002]. The pur-pose of the PCA analysis was to reveal latent structuresexisting in the data. This allowed us to identify data com-ponents that clustered together and may have a systematicrelationship. Accordingly, separate models were neededfor each PCA cluster. PCA and PLS techniques share acommon basis. Both techniques define the underlyingstructure of the data (i.e., the latent variables) by projec-ting linear planes into multidimensional data [Kramer,1998]. PLS has one important feature compared to PCA:PCA determines latent variables only in one matrix (i.e.,predictors), whereas PLS takes into account both the inde-pendent (X) and the dependent (Y) variables to build apredictive model. In our case, the PLS approach translatesinto a predictive model for Y (changes in clinical scores)depending on X (changes in power spectra). Put other-wise, the model describes which components of the powerspectra covary with changes in symptoms. Each latentvariable will explain a part of the variance in the data.Typically, several components are needed to explain mostof the variances in the data. Also, it is desirable to have asmaller number of latent variables, for the results to bemore manageable and accountable. Taking this intoaccount, the goal of the PLS regression modeling was toexplain the greatest part of the variations in the data withthe smallest number of components. PLS was used tomodel the relationship between the 595 normalized 1-Hzwide power bands as predictive variables on the one handand the groups of clinical test scores, revealed by PCA, asdependent variables on the other hand.

The goodness of fit of the model was evaluated in termsof both R2 (which is analogous to the Pearson product–moment correlation coefficient) and the goodness ofprediction Q2 as determined by a crossvalidation proce-dure [Vandervoet, 1994; Wold, 1978]. Both R2 and Q2 were

r Adamchic et al. r

r 2104 r

calculated for the original data as well as for 20 randompermutations of the dependent variable matrix Y, whereaskeeping the X matrix fix. As the matrix of dependentvariables [Y1] changes to a new, reordered matrix [Yr], thecorrelation between these two matrices (corr[Y1,Yr])decreases. If there is a nonrandom relationship betweenpredictors and dependent variable, the predictive PLSmodel, constructed on the reordered data, should have apredictive power (Q2) that decreases as the correlationbetween Y1 and Yr decreases. If this is not the case, andthe PLS model for the reordered data shows a similar pre-dictive ability (Q2) as for the initial data, there is a goodpossibility for the model being based on chance.

We excluded less informative and redundant X variables(i.e., variables with low contribution for explaining the Yvariable) from the modeling by the variable selection proce-dure [Marini et al., 2005]. For this purpose, the variable im-portance in the projection (VIP) was used that has beencomputed according to the formula used in the SIMCA-P 1

12.0.1 (http://www.umetrics.com/). Predictive variablesthat have a VIP value of >1 are most relevant for explainingdependent variables. Accordingly, all descriptors with aVIP value less than 1 were excluded from the model. Thisprocedure was iteratively repeated until an optimal model(the one with the highest Q2) was obtained.

RESULTS

Clinical Data

Mean improvement in the TQ and VAS-L=VAS-A scoresfor all 28 patients with bilateral tinnitus after 12 weeks oftreatment was 211.0 6 10.2 (P < 0.01) and 219.5 6

22.6=219.3 6 21.6 (P < 0.01=P < 0.01), respectively. Thesubgroup of good responders (n 5 12) showed a meanreduction of 219.5 6 13.5 (P < 0.001) TQ points and226.7 6 26.2=225.4 6 24.0 (P < 0.01=P < 0.01) points inVAS-L=VAS-A. At 12 weeks, PTAs in the group of all tin-nitus patients were 14.07 (10.25) for 1–4 kHz and 36.56(17.80) for 6–12 kHz. In the group of good responders at12 weeks, PTAs were 13.08 (9.25) for 1–4 kHz and 31.32(15.21) for 6–12 kHz. No significant differences were foundbetween baseline and the 12-week visit PTAs in the tinni-tus patients. Significant differences (i.e., P < 0.05) in thehearing thresholds (PTA, 6–12 kHz) were found betweenall tinnitus patients (n 5 28; both at baseline and at 12weeks) and the control group.

Altered EEG Rhythmicity

Mean power spectra

Let us first consider the EEG at baseline, that is, prior to CRtherapy. To summarize the data from all EEG scalp sensorsand all subjects, we averaged the individual spectra from alltinnitus patients recorded before CR neuromodulation (n 5

28), from the subgroup of good responders (n 5 12) and from

the healthy controls (n 5 16), respectively (Fig. 1). The averagepower spectrum in the group of all bilateral tinnitus patientsbefore CR therapy (n 5 28) showed clear differences from theaverage power spectrum of the control group. In the patientsgroup, the average spectral power was higher in d (1–3.5 Hz),low h (�5 Hz), high b (�20 Hz), low c (30.5–48 Hz), and highc (52–90 Hz) bands. In contrast, the spectral power in the higha band (9.7–12 Hz) was lower in the tinnitus patients group.In the tinnitus population, the a peak frequency (i.e., the fre-quency at which the a peak was located) prior to CR therapywas shifted toward lower frequencies as compared to thehealthy controls; however, this difference was not significant(P 5 0.11). The subgroup of 12 good responders (Fig. 1A, bluedashed line) showed the same differences from the healthypopulation in all frequency ranges as the complete group of28 bilateral patients before therapy (Fig. 1A, red line).

BESA source montage analysis

Furthermore, we studied which brain regions contrib-uted most to the significant differences between patient

Figure 1.

Enhanced d and c EEG power in tinnitus patients. A: In the

global EEG power spectrum, d and c power were increased and

a power was decreased in tinnitus patients recorded before CR

neuromodulation (n 5 28, red line) as compared to the healthy

controls (n 5 16, green line). In contrast, the global EEG power

spectrum of the group of good responders (n 5 12, blue dashed

line) did not differ from that of the group of all patients with

bilateral tinnitus before CR therapy. B: Areas indicated with red

(i.e., above or below the horizontal line) correspond to statisti-

cally significant differences between all bilateral tinnitus patients

(n 5 28) and the healthy controls. Significant differences were

found in d, low h, a, high b, low and high c bands.

r Reversing Pathological Neural Activity in Tinnitus r

r 2105 r

and control group and at which frequencies. Mann–Whitney–U–tests were performed for each regional sourceat each frequency point (Fig. 2). The maximal Z-value (jZj5 3.6) was observed in the 8–12 Hz a band in the tempo-ral regions. Z-values in the temporal region were highestcompared to the other ROIs in most of the frequencybands. The subgroup of 12 patients with a TQ improve-ment of �12 showed no significant differences in spectralpower to the subgroup of 16 patients with a TQ improve-ment of <12 in any of the ROIs: temporal (F 5 0.79, P 5

0.59), PA (F 5 0.75, P 5 0.62), DPFC (F 5 0.54, P 5 0.77),OF (F 5 0.88, P 5 0.52), anterior cingulate (F 5 1.01, P 5

0.45), and posterior cingulate (F 5 1.53, P 5 0.22). We alsoinvestigated possible effects of tinnitus duration in patientswith a tinnitus history of �4 years and >4 years, respec-tively [Schlee et al., 2009a]. There were no significant dif-ferences in spectral power between patients with tinnitusduration of �4 years and patients with tinnitus durationof >4 years in any ROIs: temporal (F 5 0.57, P 5 0.75), PA(F 5 0.62, P 5 0.71), DPFC (F 5 0.69, P 5 0.66), OF (F 5

0.65, P 5 0.69), anterior cingulate (F 5 0.75, P 5 0.62), andposterior cingulate (F 5 0.66, P 5 0.68).

Standardized low-resolution brain electromagnetictomography

sLORETA images show significant power differences indifferent frequency bands between tinnitus patients andhealthy controls (Fig. 3). As opposed to the healthy con-trols, the cortical spectral power of all bilateral tinnituspatients before the start of the treatment was significantlyenhanced in the d, b, low c, and high c frequency band. Incontrast, in the a band, we observed a reduced corticalspectral power in the tinnitus population as compared tothe healthy controls. The reduced a power was found inprefrontal dorsolateral and medial, anterior and posteriorcingulate, orbitofrontal, parietal and temporal, and otherregions (Fig. 2 and Table II).

Spatial peaks (i.e., local maxima) of the spectral powerprior to the start of the therapy were localized in the tem-poral cortex in the d band (Brodman area [BA] 42, t 5

8.98) and in the temporal and DPFC in the low c band(BA 41, t 5 6.43; BA 46, t 5 6.36). Spatial troughs (i.e.,local minima) of the spectral power were found in the aband in the temporal (BA 22, t 5 9.16) and in the DPFC(BA 46, t 5 9.23). No significant differences between tinni-tus patients and healthy controls were found in the h fre-quency range.

Treatment-Induced Changes

Furthermore, we studied the effect of acoustic CR neuro-modulation on the spontaneous oscillatory brain activity.To this end, first, we investigated the group of all patientswith bilateral tinnitus after 12 weeks of CR treatment. Thepower spectrum of this patient group approached the

average spectrum of the healthy control group, with themost prominent changes being observed in the temporalregions (Fig. 2). Consequently, we observed a decrease ofthe Z-values, indicative of significant differences betweentinnitus patients and healthy controls, in a wide range offrequencies (d, a, low and high c) after 12 weeks of CRtherapy (Fig. 2). Further, we investigated a group of goodresponders (n 5 12), that is, patients with a pronouncedreduction of their tinnitus symptoms: TQ reduction �12points. The average EEG spectrum of these patientsapproached the average spectrum of the healthy controlsto an even greater extent than that of all 28 bilateral tinni-tus patients and, accordingly, resulted in lower Z-valuesfor d, a, low and high c bands and disappearance of signif-icant differences in h and b bands (Fig. 2). No significantdifferences were found in any of the ROIs in the group of12 good responders after 12 weeks of therapy (Fig. 2).

Statistical sLORETA images reveal a reduction of thenumber of voxels and, consequently, the number of BAsdisplaying a significantly different power between all bilat-eral tinnitus patients after therapy and healthy controls(Fig. 3 and Table II). The number of voxels with powervalues that significantly differed from those of the healthypopulation was reduced in the group of all 28 tinnituspatients (Table III). To an even greater extent, the numberof voxels with significant power differences decreased inthe group of good responders after 12 weeks of CR treat-ment (Table III).

Interestingly, spatial peaks (i.e., local maxima) of thespectral power in the group of all bilateral tinnitus patientsafter 12 weeks of therapy were still localized in the tempo-ral cortex in the d (BA 41, t 5 7.53) and low c (BA 41, t 5

5.24) band. In contrast, in the group of 12 good respondersthe spatial peaks (i.e., local maxima) of the spectral powerwere located in the frontal lobe for d (BA 44, t 5 5.67) andlow c (BA 46, t 5 4.32), whereas the temporal lobes lostnearly all of their significantly different voxels after 12weeks of CR therapy (Fig. 3). Remarkably, no significantdifferences were found after treatment in the group of goodresponders in the b and high c frequency band anymore.A list of BAs displaying significant results according to thesLORETA comparison between tinnitus patients versushealthy controls is summarized in the Table II.

Association of Changes in Spontaneous Brain

Activity with Reduction of Tinnitus Symptoms

Principal component analysis

The loading plot in Figure 4 shows the results of thePCA on the changes of the VAS and TQ scores. The load-ing plot shows a clustering of the TQ and VAS accordingto the similarity of their responses to the therapy. The TQPD and TQ I (intrusiveness) subscores and the TQ totalscore turn out to cluster together, whereas VAS-L andVAS-A form another and separate cluster. The TQ A (audi-tory perceptual difficulties) and TQ Si (sleep disturbances)

r Adamchic et al. r

r 2106 r

Figure 2.

Power spectra for the group of healthy controls (n 5 16) (green

line), the group of all bilateral tinnitus patients (n 5 28) before

therapy (red line), the group of all bilateral tinnitus patients (n 5

28) after therapy (orange dashed line), and the subgroup of good

responders (n 5 12) (blue line) in the temporal (A), PA (B) and

DPFC (C), OF (M), anterior cingulated (N), and posterior cingu-

lated (O) ROIs. The power spectra of all 28 bilateral tinnitus

patients before therapy were compared to those of the healthy

controls (n 5 16) by means of the Mann–Whitney U-test in the

temporal (D), PA (E) and DPFC (F), OF (P), anterior cingulated

(Q), and posterior cingulated (R) ROIs for each frequency point.

In the same manner, the power spectra of all 28 bilateral patients

after therapy were compared to the spectra of the healthy con-

trols (n 5 16) (G, H, I, S, T, U). The analogous comparison was

also performed between the good responders (n 5 12) and the

healthy controls (n 5 16) (J, K, L, V, W, X) in the temporal, PA,

DPFC, OF, anterior cingulated, and posterior cingulated ROIs,

respectively. Areas indicated with red (i.e., above or below the

horizontal line) correspond to statistically significant differences.

In plots with neither red areas nor horizontal significance lines,

no frequency point attained significance threshold. There is a no-

ticeable trend toward a reduced a peak frequency in the tinnitus

population. Supporting Information Figure S1 shows topographical

plots of EEG power differences in the specific frequency bands.

Figure 3.

sLORETA functional tomographic maps of the significant differen-

ces in the power of regional electric brain activity between all 28

bilateral tinnitus patients before therapy and healthy controls

(“Before therapy”), between all 28 bilateral tinnitus patients after

12 weeks of therapy and healthy controls (“After therapy all tinni-

tus patients n 5 28”) and between 12 good responders after 12

weeks of therapy and healthy controls (“After therapy good res-

ponders n 5 12”) in five EEG frequency bands d (1–3.5 Hz), a(8–12 Hz), b (12.5–30 Hz), low c (30.5–48 Hz), and high c (52–

90 Hz). In the h band (4–7.5 Hz) no significant differences were

found between tinnitus patients and healthy controls, neither

before nor after CR therapy. Voxels of significantly decreased

power (P 5 0.05) in tinnitus patients as compared to healthy con-

trols are labeled in blue, whereas voxels with significantly

increased power (P 5 0.05) are labeled in red. After 12 weeks of

acoustic CR neuromodulation, there was a pronounced decrease

of number of voxels with power that was significantly different

from the healthy controls: In the d, b as well as low and high cband initially pathologically enhanced power decreased, whereas

in the a band the initially reduced power reincreased.

subscores were both located at different and, in particular,remote sites compared to the two clusters mentionedabove. This indicates that changes in TQ PD and TQ I sub-scores and the TQ total score on the one hand and VAS-Land VAS-A on the other hand as well as TQ A and TQ Sisubscores, respectively, responded differently to the ther-apy and might be associated with different underlyingmechanisms or were connected to similar mechanisms butto a different extent. Based on these results, a further PLSregression modeling was performed, TQ PD and TQ I sub-scores as well as TQ total scores were analyzed separatelyfrom TQ A and TQ Si subscores, being further separatedfrom VAS-L and VAS-A.

Partial least-squares

First, we performed the PLS for the spectral power vari-ables versus three TQ-dependent variables (TQ PD, TQ Isubscores, and TQ total scores). This resulted in a PLSmodel with good performance (R2 5 0.66, Q2 5 0.4). Amodel validation by permutations indicated the presenceof nonrandom associations in the model. The strength ofthe associations between EEG and TQ scores was deter-mined according to a loading plot (Fig. 5).

This loading plot shows the clustering of the predictorvariables and the dependent variables, indicating thestrength of the association (covariance) between changesin neuronal power with changes in the individual TQscores. The loading plot combines both PLS componentsincluded into the model. The widest positive associationwas found between the changes in TQ scores and thechanges in AC1 b, low and high c activity. Numerous fre-quency bands in other sources were also positively associ-ated with the changes in symptoms as assessed by the TQ(sub-)scores (Fig. 5). Other frequency bands (located in theleft lower quadrant of Fig. 5) were negatively associatedwith the TQ PD and TQ I subscores and the TQ totalscores.

A similar model performance was obtained for the PLSmodel with respect to changes of the VAS-L and VAS-Ascores. Two components with 67 predictors resulted in amodel with good performance (R2 5 0.65, Q2 5 0.41).Figure 6 shows the loading plot with the VAS-L and VAS-A as dependent variables and the frequency bands as pre-dictor variables.

The most pronounced positive associations with thechanges in the VAS were found for the changes in the fol-lowing sources and frequency bands: ACI and ACII highd/low h, b, low and high c. Frequency bands in the leftbottom corner of the Figure 6 were negatively associatedwith the VAS-L and VAS-A. Interestingly, in the PLSmodel for the VAS scores, as opposed to the PLS modelfor the TQ scores, we found a higher influence of limbicareas (CA and CP) on the clinical VAS scores (Fig. 6). Nosatisfactory PLS models were found for TQ A and TQ Sisubscores. No significant correlations were found betweenhearing thresholds (PTAs) and power in any of the fre-quency bands.

TABLE II. Statistically significant results from the

comparison of the group of all bilateral tinnitus patients

after 12 weeks of therapy (n 5 28) with healthy controls

and the group of good responders after 12 weeks of

therapy (n 5 12) with healthy controlsa

d Left d Right1, 2, 3, 4, 6, 8, 9, 10, 11, 13,

18, 19, 20, 21, 22, 23, 24, 27, 28,29, 30, 32, 34, 35, 36, 37, 38, 39,40, 41, 42, 43, 44, 45, 46, 47

1, 2, 3, 4, 6, 8, 9, 10, 11, 13, 17,18, 19, 20, 21, 22, 23, 24, 25,27, 28, 29, 30, 32, 33, 34, 35,36, 37, 38, 39, 40, 41, 42, 43,44, 45, 46, 47

a Left a Right1, 2, 3, 4, 6, 8, 9, 10, 11, 13,

17, 18, 19, 20, 21, 22, 23, 24, 25,27, 28, 29, 30, 31, 32, 33, 34, 35,36, 37, 38, 39, 40, 41, 42, 43, 44,45, 46, 47

1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 13,17, 18, 19, 20, 21, 22, 23, 24,25, 27, 28, 29, 30, 31, 32, 33,34, 35, 36, 37, 38, 39, 40, 41,42, 43, 44, 45, 46, 47

b Left b Right13, 22, 40, 41, 42 22, 40, 42Low c left Low c right1, 2, 3, 4, 6, 8, 9, 10, 11, 13, 19,

20, 21, 22, 27, 28, 32, 34, 35,36, 37, 38, 39, 40, 41, 42, 43,44, 45, 46, 47

1, 2, 3, 4, 6, 8, 9, 10, 11, 13, 18,19, 20, 21, 22, 23, 25, 27, 28,29, 30, 32, 34, 35, 36, 37, 38,40, 41, 42, 43, 44, 45, 46, 47

High c left13, 22, 40, 41, 42

High c right13, 20, 21, 22, 38, 41, 42

aBAs comprising at least one significant voxel were included inthe table as areas showing significant power differences. Areasnot listed here did not show significant power differencesbetween the compared groups. BAs printed in black exhibitedsignificant differences in the group of all bilateral tinnitus patients(before and after therapy) as well as in the group of good res-ponders as compared to the healthy controls, respectively. Thepower in the black BA was, hence, significantly different from thehealthy controls and was not normalized during therapy. BAprinted in red were significantly different in the bilateral tinnituspopulation (n 5 28) from the healthy controls only before the startof the therapy. The power in the red BAs was, thus, significantlynormalized in all bilateral patients. BAs printed in blue weresignificantly different in the group of all bilateral tinnitus patients(n 5 28) from the healthy controls both before and after thetherapy, but lost their significance in the group of good respond-ers after 12 weeks of therapy (n 5 12). Hence, the power in theblue BAs was normalized after 12 weeks of therapy in the groupof good responders only.

TABLE III. Reduction of number of voxels showing a

significant power difference compared to the

healthy controls

All 28patients (%)

12 goodresponders (%)

d 242.6 277.4a 231.1 268.3b 244.9 2100.0Low c 244.4 278.8High c 283.4 2100.0

r Reversing Pathological Neural Activity in Tinnitus r

r 2109 r

DISCUSSION

We investigated oscillatory brain activity in patientswith chronic subjective tonal bilateral tinnitus before and

after 12 weeks of treatment with acoustic CR neuromodu-lation in comparison to healthy controls. To crosscheckour results, we used two qualitatively different inversetechniques: BESA source montage analysis [Scherg et al.,

Figure 4.

Loading plot showing the grouping of the clinical scores. Items grouped together responded in a

similar way to the therapy, and their proximity reflects the strength of the similarity of their

responses. The separation of TQ and VAS scores in distinct clusters suggests that they

responded differently to the therapy. TQ subscores: TQ PD (psychological distress), TQ I (intru-

siveness), TQ A (auditory perceptual difficulties), and TQ Si (sleep disturbances).

Figure 5.

Loading plot showing the changes of TQ scores and spectral

power bands for the first and second PLS component. The load-

ing plot shows the 70 1-Hz wide power bins and the TQ PD

and TQ I subscores together with the TQ total scores. The

proximity of the changes in spectral power values and the TQ

(sub-)scores reflects the strength of the association of these

changes to the changes in the clinical scores. The 1-Hz-wide

frequency bands were labeled by abbreviations consisting of the

lower edge of the frequency band followed by the ROI (for

abbreviations, see METHODS section). ROI abbreviations: pri-

mary auditory cortex (AC1), secondary auditory cortex (AC2),

orbito-frontal (OF), dorsolateral-prefrontal (DPFC), parietal

(PA), anterior cingulate cortex (CA) and posterior cingulated

cortex (CP).

r Adamchic et al. r

r 2110 r

2002] and current source density reconstruction by sLOR-ETA [Pascual-Marqui, 2002]. Overall, we found that tinni-tus patients significantly deviated from the healthycontrols, concerning their oscillatory brain activity. Fur-thermore, CR therapy significantly normalized thepatients’ brain oscillations in all frequency bands tested,even leading to a complete abolishment of pathologicalpower in several brain regions and frequency bands in thegood responders. For instance, based on the BESA sourcemontage analysis [Scherg et al., 2002], the abnormal poweralterations completely vanished in the good respondersover the entire frequency range tested (i.e., from 1 to 90Hz) in both the PA source and the cingulate source (Fig.2). With the spatially more refined sLORETA analysis[Pascual-Marqui, 2002], we revealed that in the good res-ponders pathologically enhanced b and high c power van-ished in the entire cortex, whereas d and low c decreased(nearly) completely in temporal and PA regions, whereasa power undergoes a widespread restitution (Fig. 3). Inaddition, in the patient group, the dominant a peak tendedto be shifted to lower frequencies compared to the healthycontrols.

Our results obtained with the two different inverse tech-niques, BESA source montage analysis [Scherg et al., 2002]and sLORETA analysis [Pascual-Marqui, 2002], are ingood agreement. However, in contrast to the BESA analy-sis, the sLORETA analysis did not reveal any statisticaldifferences in h band between healthy controls and tinni-tus patients. This might be owing to the methodologicaldifferences between two methods: (1) Different resolution

in the frequency domain: In contrast to our BESA analysis,where single-frequency bins were investigated, our sLOR-ETA analysis was performed on frequency bands of awidth of several hertz. (2) Different statistical methodsemployed: In our BESA analysis FDR correction for multi-ple comparisons was performed for each frequency point[Benjamini and Hochberg, 1995]. For the sLORETA analy-sis, a voxel-by-voxel comparison was fed into a nonpara-metric test for functional brain imaging that corrects formultiple comparisons [Nichols and Holmes, 2001]. (3) Dif-ferent technical features inherent to the different nature ofthe two inverse techniques, BESA [Scherg et al., 2002]being a source montage analysis as opposed to the distrib-uted current density reconstruction by sLORETA [Pascual-Marqui, 2002]. By the same token, changes of the h power,found in our previous study by comparing EEG record-ings in good responders pre- versus post-CR treatment[Tass et al., 2012a], might not have been detected by thesLORETA analysis used in this study owing to fundamen-tal methodological differences between the twoapproaches: In our previous study [Tass et al., 2012a],power changes of brain oscillations after CR neuromodula-tion were investigated in a paired groups design. In con-trast, in this study, we used an independent group designthat allowed us to compare tinnitus patients with healthycontrols combining a cross-sectional and a longitudinalapproach. On the one hand, our results are relevant froma clinical standpoint. They show electrophysiologicaleffects of acoustic CR neuromodulation and help to assessand possibly to further improve our therapeutic approach.

Figure 6.

Loading plot showing the 67 1-Hz wide power bands and VAS loudness and VAS annoyance

scores. Proximity of the changes in the power values across different frequency bands to the

changes in VAS reflects the strength of association of these neuronal changes to the changes in

the clinical scores. For ROI abbreviations, see METHODS section and Figure 5.

r Reversing Pathological Neural Activity in Tinnitus r

r 2111 r

On the other hand, a clinically successful intervention ena-bles to elucidate the pathophysiology of tinnitus, in partic-ular, the electrophysiological correlate of tinnitus (see alsothe discussion by Langguth et al. (2012)).

Electrophysiological Characteristics

of Tinnitus Patients

The enhanced synchrony at low and high frequenciescombined with the decrease of a power (Figs. 2 and 3) isconsistent with the previous findings and hence furtherindicates that this electrophysiological pattern to be a hall-mark characteristic underlying chronic subjective tinnitus.The increased power at lower and higher frequencies inour patients is in line with the previous studies, reportingsimilar spectral abnormalities in individuals with tinnitus[Adjamian et al., 2012; De Ridder et al., 2011b; Weiszet al., 2005]. Our results are at variance with a study thatrevealed increased power in the frequency range of 2–100Hz under eyes-closed and eyes-open condition in tinnituspatients as opposed to healthy controls [Moazami-Gou-darzi et al., 2010]. This might be owing to: (1) Differentcharacteristics of the analyzed groups (e.g., groups sizeand type of tinnitus); (2) Different methods used to nor-malize power spectra.

In our data, the reduced a power was not caused artifi-cially by the normalization, because in the group of tinni-tus patients the a peak is much less prominent, wider, andshifted toward lower frequencies. Furthermore, weobserve a similar spectral pattern also without the normal-ization. Our results are also in line with the study by DeRidder et al. [2011b] who described a case of a tinnituspatient with local field potential recordings via epicorticalelectrodes from the secondary auditory cortex. The patientwas recorded in states with loud as well as little to no tin-nitus. Decreased a power and increased power at lowerfrequencies were observed only in the presence of strongtinnitus [De Ridder et al., 2011b].

Increased power in spontaneous electrical brain activitywas found in chronic tinnitus patients as compared to tin-nitus patients with recent onset in h, b, and c bands [Van-neste et al., 2011b]. The chronic tinnitus patients alsodisplayed a different distribution of the c network [Schleeet al., 2009a]. However, in our study, there were no signifi-cant differences in oscillatory brain activity betweenpatients with recent tinnitus onset (�4 years) and patientswith a long history of tinnitus (>4 years). These differen-ces may be owing to different patient selection criteria orthe choice of specific frequency band boundaries selectedfor statistical analysis. We may, thus, assume that the ma-jority of the observed EEG changes in our study arerelated to the chronic tonal tinnitus, in general; however,we do not exclude the possibility of a partial effect owingto the tinnitus duration.

In our study, in the d and low c bands, that is, the bandsthat showed the most prominent and widest enhancement,

the cortical generators of excess EEG power were predom-inantly located in temporal and frontal areas with spatialpeaks of the power localized in the temporal cortex in thed band (BA 42) and in the temporal and DPFCs in the lowc band (BA 41, 46). We found the foci of altered spontane-ous oscillatory activity in tinnitus patients versus controlsto be bilaterally located in temporal areas (see above),which resembles the previous findings [Weisz et al., 2005].Concerning the differences between the symmetric bilat-eral distribution in both frontal lobes found here, asopposed to the left- or right-sided differences reported inthe previous studies, one has to take into account the rela-tively large and homogeneous group of tinnitus patientsinvestigated in our study (i.e., 28 patients with chronicbilateral tonal tinnitus) in contrast to the smaller numberof patients and=or mixed groups of patients with respectto laterality (i.e., bilateral vs. unilateral tinnitus) investi-gated in the previous studies [Moazami-Goudarzi et al.,2010; Weisz et al., 2005]. Laterality of tinnitus has beenpreviously reported to affect the pattern of spontaneousbrain activity [van der Loo et al., 2009; Vanneste et al.,2011a]. Accordingly, confinement of the analyses presentedin this manuscript to patients with bilateral tinnitus wasdone to reduce the heterogeneity which is present in tinni-tus samples. Furthermore, our finding supports the notionof Weisz et al. [2005], who hypothesized that the asymme-try between hemispheres vanishes in case groups wereused for analysis that are more balanced with respect totinnitus side. Future studies will be necessary to investi-gate if these findings can be generalized to patients withunilateral tinnitus.

Hearing Impairment

In our study, the tinnitus group had elevated auditorythresholds in the high-frequency range compared to thecontrol group. However, no significant changes in thehearing thresholds were detected in the tinnitus patientsin the pre- versus post-treatment condition. This is rele-vant in the following context. A major limitation of thepioneering MEG study by Weisz et al. [2005] was the sig-nificant difference in the status of hearing between tinnituspatients and healthy controls. That study revealed a pat-tern of brain rhythms characteristic of subjective tinnitus(enhanced d and decreased a). However, the causal speci-ficity of these findings remained a matter of debate,because in contrast to the normally hearing controls thetinnitus group had a high-frequency hearing loss [Weiszet al., 2005]. Accordingly, it was not clear whether theobserved pattern of brain rhythms had to be assigned tothe tinnitus as opposed to the hearing impairment-inducedsensory deprivation [Weisz et al., 2005]. The latter mightgive rise to similar changes of brain activity as observedduring slow-wave sleep, where brain regions of increasedd and decreased a spatially coincide [Benoit et al., 2000].However, in a previous study [Tass et al., 2012a] in one

r Adamchic et al. r

r 2112 r

group of patients with tinnitus and after significant CR-induced tinnitus relief, having the same hearing levelsbefore and after therapy, we confirmed the characteristictinnitus-related pattern of brain rhythms as reported ear-lier by Weisz et al. (2005, 2007b). This fact underlines thatthe tinnitus-related changes of brain rhythms, that is,enhanced d and c combined with decreased a [Weiszet al., 2005, 2007b], are tinnitus-specific alterations [Tasset al., 2012a]. This notion is supported by the findings ofAdjamian et al. (2012), who reported that increased d ac-tivity in tinnitus patients was likely associated with tinni-tus itself rather than deafferentation alone. Furthermore,this result is in accordance, for example, with the findingsobtained in a study comparing spontaneous EEG in tinni-tus patients with and without transitory changes of tinni-tus loudness caused by residual inhibition [Kahlbrock andWeisz, 2008]. It is important to note that the EEG record-ings analyzed in this article were obtained in our previousstudy mentioned above [Tass et al., 2012a].

Thalamocortical Dysrhythmia

Our findings are in line with the concept of the thalamo-cortical dysrhythmia (TCD) [Llinas et al., 1999; Weiszet al., 2007a], that is, a mechanism characterized by a deaf-ferentation-induced increase of thalamic and, in turn, corti-cal low-frequency oscillations, combined with corticalhigh-frequency oscillations. Along the lines of this concept,slow-wave oscillations are assumed to facilitate and sus-tain c oscillations through a mechanism called the “edgeeffect” [Llinas et al., 1999, 2005]. According to the TCDconcept, the enhanced synchronization in the c range intemporal regions (BA 41) may be regarded as a direct elec-trophysiological correlate of the auditory phantom sensa-tion [De Ridder et al., 2011b; Llinas et al., 1999; Weiszet al., 2007b]. As an indirect evidence of the validity of theTCD concept and the “edge effect,” in our tinnitus patientswe observed enhanced synchronization in low- and high-frequency bands in spatially greatly overlapping regionsof the cortex (Fig. 3). The importance of c oscillations foracoustic perception is well documented for healthy sub-jects [Crone et al., 2001] as well as for tinnitus patients [DeRidder et al., 2011b; van der Loo et al., 2009; Weisz et al.,2007b]. However, the emergence of enhanced d and c oscil-lations in frontal areas underlines the dysrhythmic coin-volvement of associative areas in the pathogenesis oftinnitus and may directly be related to the appearance oftinnitus-related affective manifestations. The involvementof associative areas in relation to emotional factors in tinni-tus has previously been reported in several studies [DeRidder et al., 2011a; Jastreboff, 1990; Lockwood et al., 1998;Mirz et al., 2000]. Additional evidence for the importanceof associative areas for a conscious phantom perceptcomes from recordings from the somatosensory and theauditory system in nonhuman primates, which indicatethat an activation of the primary sensory cortex is not

sufficient for the generation of a conscious percept and im-plicate an involvement of the association cortex in con-scious perception [Lemus et al., 2009a,2009b]. Accordingly,Jastreboff considered the prefrontal cortex as a “candidatefor the integration of sensory and emotional aspects oftinnitus” [Jastreboff, 1990]. The relevance of a cognitive–emotional network in tinnitus was also convincinglypointed out by recent studies, demonstrating abnormallong-range coupling in tinnitus patients [Schlee et al.,2009a, 2009b; Vanneste et al., 2011b]. Although the TCDconcept [Llinas and Steriade, 2006; Llinas et al., 2005, 1999]underlines the importance of pathological low- and high-frequency oscillations, the impact of reduced a oscillationsshould not be neglected [Weisz et al., 2007a]. Low levelsof a are associated with a state of excitation, whereas highlevels of a are associated with a state of inhibition [Kli-mesch et al., 2007; Weisz et al., 2007a].

Effect of CR Neuromodulation

Both groups of tinnitus patients, that is, all bilateral tin-nitus patients (n 5 28) and the subgroup of good respond-ers (n 5 12) underwent a significant reduction of tinnitussymptoms (RESULTS). In general, acoustic CR neuromo-dulation reversed the increased cortical synchronization atlow (d and low h) and high (b, low and high c) frequen-cies, observed in individuals with tinnitus at baseline, to-ward physiological levels observed in the tinnitus-freecontrol group. This finding, combined with the significantcovariance of changes in tinnitus symptoms and changesin EEG characteristics, provides further support for the hy-pothesis that tinnitus has an electrophysiological correlateas put forward previously [Llinas et al., 1999; Weisz et al.,2005].

Our results show that the activity in the slow and fastEEG frequency bands was decreased and in the a bandincreased after 12 weeks of CR neuromodulation in bothgroups with more prominent changes observed in patientswith greater tinnitus reduction. There was a significant co-variance between the degree of synchronization change in1-Hz-wide frequency bands and the degree of tinnitussymptoms change that was strongest in temporal regions.This is consistent with the previous findings, demonstrat-ing the leading role of the temporal cortex in coding tinni-tus [De Ridder et al., 2011b; van der Loo et al., 2009;Weisz et al., 2007b], especially in its early phase [Schleeet al., 2009a]. However, the function of the primary sen-sory cortices is mainly to generate an appropriate neuralrepresentation of sensory input that by itself does not leadto conscious perception [Lemus et al., 2009a]. In fact, itrequires the involvement of frontal and parietal areas, tomake such a perception become conscious [De Ridderet al., 2011a; Lemus et al., 2009b]. Strong covariance ofabnormal neuronal synchrony in the dorsolateral prefron-tal regions with changes of tinnitus distress (assessed bythe TQ), as revealed by PLS analysis, suggests that the

r Reversing Pathological Neural Activity in Tinnitus r

r 2113 r

frontal cortex is involved in a tinnitus-related cortical net-work and is more associated with the affective distress oftinnitus. Analogously, abnormal neuronal synchrony intemporal regions is strongly associated with perceptualissues (i.e., aspects concerning the loudness and characterof the sound).

Interestingly, we observed a higher covariance of thechanges in limbic areas with changes in VAS rather thanwith changes in TQ scores (Figs. 5 and 6). Why changes inoscillatory brain activity in limbic areas were morestrongly associated with VAS scores, but not with TQscores? TQ covers a relatively wide spectrum of symptomsrelated to and caused by tinnitus, referring to a longertime interval. In contrast, to a certain extent, VAS is anemotionally more biased instrument of tinnitus assess-ment. Thus, the VAS scores might be more strongly influ-enced by the momentary emotional state of the patientand, consequently, be more associated with the oscillatoryactivity in the limbic areas present during the VAS assess-ment. Previous studies indicate that tinnitus intensitydepends on the extent of abnormal neuronal synchrony ina distributed neuronal network, rather than on local coscillations [Schlee et al., 2009a; Vanneste et al., 2011b]. Inline with this notion, our results show that acoustic CRneuromodulation causes a long-lasting and widespreaddecrease of the power of both slow- and fast-brain oscilla-tions in a large network comprising auditory and nonaudi-tory areas. In fact, this pronounced normalization of brainoscillations together with the significant decrease of thetinnitus frequency [Tass et al., 2012a] is indicative of plas-tic changes induced by the CR therapy.

Comparison to Neurofeedback Training

Neurofeedback training is based on operant learningand aims at a mentally induced reinforcement of beneficialphysiological processes mediated by a feedback of appro-priate physiological variables [Sterman and Friar, 1972].Initially, neurofeedback training in the treatment of tinni-tus has been applied mainly with the aim to induce relaxa-tion by upregulating posterior a [Gosepath et al., 2001;Schenk et al., 2005]. In that context, it has been suggestedthat distress generally is associated with a reduction ofposterior a power and an enhancement of b power (14–30Hz) [Gosepath et al., 2001; Schenk et al., 2005]. An upregu-lation of the a activity might, thus, be interpreted as a signof increased relaxation. As a next step, regionally specific,presumably tinnitus-related brain activity was used asfeedback [Dohrmann et al., 2007a,b]: Instead of posteriorEEG sites, Dohrmann et al. (2007a,b) used frontocentralEEG sites (F3, F4, FC1, and FC2, referenced to the rightmastoid) as electrical generators located in the auditorycortex project to these frontal sites [Pantev et al., 1995].

Two different types of neurofeedback training with fron-tocentral EEG signals were used to attenuate tinnitus: a–dtraining aims at a normalization of the spontaneous EEG

by increasing a and decreasing d power [Dohrmann et al.,2007a,b; Weisz et al., 2011]. In contrast, the goal ofdesynchronization suppression training is to maximize apower during sound stimulation and, hence, to mentallycounteract the “natural” tendency of a desynchronization[Weisz et al., 2011]. However, both a–d training [Dohr-mann et al., 2007a,2007b; Weisz et al., 2011] and desynch-ronization suppression training [Weisz et al., 2011] lead toonly a marginal modulation of d power [Weisz et al.,2011]. Accordingly, it is still a matter of debate whetherneurofeedback training specifically modulates tinnitus-related brain activity, or rather simply affects nonspecificneuronal processes that influence tinnitus distress [Weiszet al., 2011]. Given our results of the comparison to base-line (i.e., prior to therapy; Tass et al., 2012a) as well as thecomparison to healthy controls presented here, acousticCR neuromodulation, in fact, specifically reverses the tin-nitus characteristic EEG patterns by decreasing d and cand increasing a power in a tinnitus-related network com-prising auditory and nonauditory brain areas.

Mechanism of CR Neuromodulation

In this study and in a previous study [Tass et al., 2012a],we analyzed the amount of neuronal synchronization (interms of spectral power in EEG signals) before and afteracoustic CR neuromodulation. In particular, we did notinvestigate the EEG dynamics during CR neuromodula-tion. Accordingly, as yet, we cannot determine in whichway acoustic CR neuromodulation causes the long-lastingdesynchronization in widespread cortical neuronal popula-tions. Based on the previous, in particular, computationalstudies, CR neuromodulation might cause a desynchroni-zation of cortical neuronal populations in qualitatively dif-ferent ways [Popovych and Tass, 2012; Tass andPopovych, 2012]. Given the tonotopic organization of thecentral auditory system [Ehret and Romand, 1997], andowing to the fact that phase resets may propagate via syn-apses [Jackson et al., 2002; Lerma and Garciaaustt, 1985;Perkel et al., 1964; Pinsker, 1977; Popovych and Tass, 2012;Prinz et al., 2003; Tass and Popovych, 2012], stimulation-induced phase resets may, hence, occur at different hier-archical levels of the central auditory system. The tono-topic selectivity, that is, the spatial spread of the differentstimuli crucially depends on the characteristics of the tun-ing curves of auditory nerve fibers, which may, for exam-ple, be pathologically broadened owing to a cochlearhearing loss [Givens, 1996; Perkel et al., 1964; Prinz et al.,2003; Ryan et al., 1979]. However, as shown computation-ally, CR neuromodulation is robust against variations ofthe spatial spread [Lysyansky et al., 2011]. For this reason,acoustic CR neuromodulation might, in principle, induce adesynchronization of the d rhythm at the level of the pri-mary auditory cortex by causing phase resets of the drhythm in different cortical subpopulations at differenttimes. However, CR might cause a desynchronization at

r Adamchic et al. r

r 2114 r

the cortical level also in a qualitatively different way. CRmight cause a desynchronization of an upstream nucleusby its typical mechanism, that is, via time-shifted phaseresets of neuronal subpopulations [Tass, 2003], whereasfurther downstream in the central auditory systemdesynchronization might be achieved in a, so to speak,unspecific manner, that is, by a downstream propagationof the desynchronizing effect. Computationally it has beenshown that desynchronizing effects may propagate fromone neuronal population to another [Hauptmann et al.,2005; Popovych and Tass, 2010]. The “unspecific” propaga-tion mechanism of desynchronization might be particu-larly relevant for the propagation of desynchronizingeffects from auditory to nonauditory areas as this mecha-nism does not require a tonotopic organization of the con-nections. It is important to note that prior to CR therapyspatial peaks (i.e., local maxima) of the spectral power inthe group of all bilateral tinnitus, patients were localizedin the primary auditory cortex (BA 41) in the d and low cband (Fig. 3). In contrast, in the group of 12 good respond-ers, the spatial peaks of the spectral power were found inthe frontal lobe for d (BA 44) and low c (BA 46), whereasin the temporal lobes nearly all of the significantly differ-ent voxels vanished after 12 weeks of CR therapy(RESULTS and Fig. 3). This remarkable finding might beowing to the fact that the CR-induced desynchronizationfirst reaches the primary auditory cortex from where itpropagates further downstream, into the tinnitus-relatednetwork comprising different, and especially nonauditorycortical areas. A propagation process of this kind, whichinvolves a larger network of neuronal populations, willlikely take time, possibly in a similar way as it is knownfrom, for example, propagation phenomena in oscillatoryor excitable media [Mikhailov, 1990]. Accordingly, a com-plete evolution of this desynchronization propagationmight, on average, require more than 12 weeks of CRtherapy.

High levels of test retest reliability of quantitative EEG(e.g., LORETA) were demonstrated over many days andweeks [Cannon et al., 2012; Thatcher, 2010]. However, hav-ing performed only a single EEG recording in every sub-ject from the control group might be considered as alimiting factor of this study as test–retest reliability ofthese recordings was not assessed. Accordingly, in futurestudies it may be desirable to assess the reliability of CR-induced EEG normalization, observed in this study, usingBESA source montage and sLORETA techniques to sub-stantiate their role as an indicator of the course of tinnitustreatment.

CONCLUSIONS

In this study, we focused on CR-induced EEG powermodulations and their association to the changes of thetinnitus sensation. CR-induced modulation of the effectiveconnectivity in a tinnitus-related network, comprising

auditory and nonauditory brain areas, was beyond thescope of this manuscript and will be investigated in a sep-arate study [Silchenko et al., 2013]. In this study, we didnot investigate the EEG dynamics during CR neuromodu-lation. Thus, as yet, we cannot determine the specific waythat acoustic CR leads to a normalization of the oscillatorypower in the tinnitus-related network of brain areas. Fur-ther studies are necessary to further investigate the specificmechanisms of CR treatment and its comparison to otherauditory stimulation paradigms. Accordingly, anotherstudy will be devoted to the dynamical mechanism ofacoustic CR neuromodulation in tinnitus patients. Toreveal the actual mechanism by which CR causes adesynchronization on the cortical level, we shall investi-gate in which auditory and nonauditory brain areas singletones cause a phase reset of pathological rhythms, forexample, the d rhythm, whereas the corresponding CRstimuli, that is, the time-shifted sequences of phase reset-ting tones, cause a desynchronization. Also, further com-putational studies will be devoted to the propagation ofdesynchronizing effects in networks of interacting neuro-nal populations with different connection topologies topossibly optimize the stimulation protocol, for example, byidentifying stimulation parameters that, properly chosen,might enable to substantially increase the propagation ve-locity. The goal of this combined theoretical and experi-mental approach is to optimize the CR therapy in a waythat also the, fortunately, smaller portion of patients whostill do not sufficiently profit from the current CR versionmight finally benefit as well.

CONFLICT OF INTEREST

I. Adamchic and T. Toth have no conflict of interest inrelation to this study. C. Hauptmann and P. A. Tass havea contractual relationship with ANM Adaptive Neuromo-dulation GmbH. P. A. Tass holds shares of ANM AdaptiveNeuromodulation GmbH.

REFERENCES

Adamchic I, Langguth B, Hauptmann C, Tass PA (2012a): Psycho-metric evaluation of Visual Analog Scale for the assessment ofchronic tinnitus. Am J Audiol 21:215–225.

Adamchic I, Tass PA, Langguth B, Hauptmann C, Koller M,Schecklmann M, Zeman F, Landgrebe M (2012b): Linking theTinnitus Questionnaire and the subjective Clinical GlobalImpression: Which differences are clinically important? HealthQual Life Outcomes 10:79.

Adjamian P, Sereda M, Zobay O, Hall DA, Palmer AR (2012):Neuromagnetic indicators of tinnitus and tinnitus masking inpatients with and without hearing loss. J Assoc Res Otolaryn-gol 13:715–731.

Benjamini Y, Hochberg Y (1995): Controlling the False DiscoveryRate: A practical and powerful approach to multiple testing. JR Stat Soc Ser B 57:289–300.

r Reversing Pathological Neural Activity in Tinnitus r

r 2115 r

Benoit O, Daurat A, Prado J (2000): Slow (0.7–2 Hz) and fast (2–4Hz) delta components are differently correlated to theta, alphaand beta frequency bands during NREM sleep. Clin Neuro-physiol 111:2103–2106.

Cannon RL, Baldwin DR, Shaw TL, Diloreto DJ, Phillips ST,Scruggs AK, Riehl TC (2012): Reliability of Quantitative EEG(qEEG) measures and LORETA current source density at 30days. Neurosci Lett 518:27–31.

Crone NE, Boatman D, Gordon B, Hao L (2001): Induced electro-corticographic gamma activity during auditory perception.Brazier Award-winning article, 2001. Clin Neurophysiol112:565–582.

Delorme A, Sejnowski T, Makeig S (2007): Enhanced detection ofartifacts in EEG data using higher-order statistics and inde-pendent component analysis. Neuroimage 34:1443–1449.

De Ridder D, De Mulder G, Menovsky T, Sunaert S, Kovacs S(2007): Electrical stimulation of auditory and somatosensorycortices for treatment of tinnitus and pain. Tinnitus: Pathophy-siol Treat 166:377–388.

De Ridder D, Elgoyhen AB, Romo R, Langguth B (2011a): Phan-tom percepts: Tinnitus and pain as persisting aversive memorynetworks. Proc Natl Acad Sci USA 108:8075–8080.

De Ridder D, van der Loo E, Vanneste S, Gais S, Plazier M,Kovacs S, Sunaert S, Menovsky T, van de Heyning P (2011b):Theta-gamma dysrhythmia and auditory phantom perception.J Neurosurg 114:912–921.

Dietrich V, Nieschalk M, Stoll W, Rajan R, Pantev C (2001): Corti-cal reorganization in patients with high frequency cochlearhearing loss. Hear Res 158:95–101.

Dohrmann K, Elbert T, Schlee W, Weisz N (2007a): Tuning the tin-nitus percept by modification of synchronous brain activity.Restor Neurol Neurosci 25:371–378.

Dohrmann K, Weisz N, Schlee W, Hartmann T, Elbert T (2007b):Neurofeedback for treating tinnitus. 166:473–554.

Eggermont JJ, Roberts LE (2004): The neuroscience of tinnitus.Trends Neurosci 27:676–682.

Ehret G, Romand R (1997): The Central Auditory System. NewYork: Oxford University Press.

Gerstner W, Kempter R, vanHemmen JL, Wagner H (1996): Aneuronal learning rule for sub-millisecond temporal coding.Nature 383:76–78.

Gopinath B, McMahon CM, Rochtchina E, Karpa MJ, Mitchell P(2010): Risk factors and impacts of incident tinnitus in olderadults. Ann Epidemiol 20:129–135.

Gosepath K, Nafe B, Ziegler E, Mann WJ (2001): [Neurofeedbackin therapy of tinnitus]. HNO 49:29–35.

Hauptmann C, Tass PA (2007): Therapeutic rewiring by means ofdesynchronizing brain stimulation. Biosystems 89:173–181.

Hauptmann C, Popovych O, Tass PA (2005): Effectively desynch-ronizing deep brain stimulation based on a coordinateddelayed feedback stimulation via several sites: A computa-tional study. Biol Cybern 93:463–470.

Hegerl U, Gallinat J, Mrowinski D. 1994: Intensity dependence ofauditory-evoked dipole source activity. Int J Psychophysiol17:1–13.

Ille N, Berg P, Scherg M (2002): Artifact correction of the ongoingEEG using spatial filters based on artifact and brain signal top-ographies. J Clin Neurophysiol 19:113–124.

Irvine DR, Rajan R, Brown M (2001): Injury- and use-related plas-ticity in adult auditory cortex. Audiol Neurootol 6:192–195.

Jackson A, Spinks RL, Freeman TCB, Wolpert DM, Lemon RN(2002): Rhythm generation in monkey motor cortex exploredusing pyramidal tract stimulation. J Physiol 541:685–699.

Jacobson NS, Truax P (1991): Clinical significance: A statisticalapproach to defining meaningful change in psychotherapyresearch. J Consult Clin Psychol 59:12–19.

Jastreboff PJ (1990): Phantom auditory perception (tinnitus): Mech-anisms of generation and perception. Neurosci Res 8:221–254.

John ER, Prichep L, Ahn H, Easton P, Fridman J, Kaye H (1983):Neurometric evaluation of cognitive dysfunctions and neuro-logical disorders in children. Prog Neurobiol 21:239–290.

Jolliffe IT (2002): Principal Component Analysis. New York:Springer.

Kahlbrock N, Weisz N (2008): Transient reduction of tinnitus in-tensity is marked by concomitant reductions of delta bandpower. Biomed Chromatogr Biol 6:4.

Klimesch W, Sauseng P, Hanslmayr S (2007): EEG alpha oscilla-tions: The inhibition-timing hypothesis. Brain Res Rev 53:63–88.

Kramer R (1998): Chemometric Techniques for Quantitative Anal-ysis. Boca Raton: Taylor & Francis.

Krishnan A, Williams LJ, McIntosh AR, Abdi H (2011): PartialLeast Squares (PLS) methods for neuroimaging: A tutorial andreview. Neuroimage 56:455–475.

Langguth B, Landgrebe M, Kleinjung T, Sand GP, Hajak G (2011):Tinnitus and depression. World J Biol Psychiatry 12:489–500.

Langguth B, Schecklmann M, Lehner A, Landgrebe M, Poeppl TB,Kreuzer PM, Schlee W, Weisz N, Vanneste S, De Ridder D(2012): Neuroimaging and neuromodulation: Complementaryapproaches for identifying the neuronal correlates of tinnitus.Front Syst Neurosci 6:15.

Lanting CP, de Kleine E, van Dijk P (2009): Neural activity under-lying tinnitus generation: Results from PET and fMRI. HearRes 255:1–13.

Lemus L, Hernandez A, Romo R (2009a): Neural codes for percep-tual discrimination of acoustic flutter in the primate auditorycortex. Proc Natl Acad Sci USA 106:9471–9476.

Lemus L, Hernandez A, Romo R (2009b): Neural encoding of au-ditory discrimination in ventral premotor cortex. Proc NatlAcad Sci USA 106:14640–14645.

Lerma J, Garciaaustt E (1985): Hippocampal theta rhythm duringparadoxical sleep—Effects of afferent stimuli and phase-rela-tionships with phasic events. Electroencephalogr Clinl Neuro-physiol 60:46–54.

Llinas RR, Steriade M (2006): Bursting of thalamic neurons andstates of vigilance. J Neurophysiol 95:3297–3308.

Llinas RR, Ribary U, Jeanmonod D, Kronberg E, Mitra PP (1999):Thalamocortical dysrhythmia: A neurological and neuropsychi-atric syndrome characterized by magnetoencephalography.Proc Natl Acad Sci USA 96:15222–15227.

Llinas R, Urbano FJ, Leznik E, Ramirez RR, van Marle HJ (2005):Rhythmic and dysrhythmic thalamocortical dynamics: GABAsystems and the edge effect. Trends Neurosci 28:325–333.

Lockwood AH, Salvi RJ, Coad ML, Towsley ML, Wack DS, Mur-phy BW (1998): The functional neuroanatomy of tinnitus: Evi-dence for limbic system links and neural plasticity. Neurology50:114–120.

Marini F, Roncaglioni A, Novic M (2005): Variable selection andinterpretation in structure-affinity correlation modeling ofestrogen receptor binders. J Chem Inf Model 45:1507–1519.

r Adamchic et al. r

r 2116 r

Markram H, Lubke J, Frotscher M, Sakmann B (1997): Regulationof synaptic efficacy by coincidence of postsynaptic APs andEPSPs. Science 275:213–215.

McIntosh AR, Bookstein F, Haxby J, Grady C (1996): Spatial pat-tern analysis of functional brain images using partial leastsquares. Neuroimage 3:143–157.

Merzenich MM, Nelson RJ, Stryker MP, Cynader MS, Schopp-mann A, Zook JM (1984): Somatosensory cortical map changesfollowing digit amputation in adult monkeys. J Comp Neurol224:591–605.

Mikhailov AS (1990): Foundations of Synergetics: DistributedActive Systems. Berlin: Springer.

Mirz F, Gjedde A, Ishizu K, Pedersen CB (2000): Cortical networkssubserving the perception of tinnitus—A PET study. Acta Oto-laryngol Suppl 543:241–243.

Moazami-Goudarzi M, Michels L, Weisz N, Jeanmonod D (2010):Temporo-insular enhancement of EEG low and high frequen-cies in patients with chronic tinnitus. QEEG study of chronictinnitus patients. Biomed Chromatogr Neurosci 11:40.

Nichols TE, Holmes AP (2001): Nonparametric permutation testsfor functional neuroimaging: A primer with examples. HumBrain Mapp 15:1–25.

Norena A, Micheyl C, Chery-Croze S, Collet L (2002): Psycho-acoustic characterization of the tinnitus spectrum: Implicationsfor the underlying mechanisms of tinnitus. Audiol Neurootol7:358–369.

Ochi K, Eggermont JJ (1997): Effects of quinine on neural activityin cat primary auditory cortex. Hear Res 105:105–118.

Pantev C, Bertrand O, Eulitz C, Verkindt C, Hampson S, SchuiererG, Elbert T (1995): Specific tonotopic organizations of differentareas of the human auditory-cortex revealed by simultaneousmagnetic and electric recordings. Electroencephalogr Clin Neu-rophysiol 94:26–40.

Pascual-Marqui RD (2002): Standardized low-resolution brain elec-tromagnetic tomography (sLORETA): Technical details. Meth-ods Find Exp Clin Pharmacol 24:5–12.

Perkel DH, Schulman JH, Segundo JP, Bullock TH, Moore GP(1964): Pacemaker neurons—Effects of regularly spaced synap-tic input. Science 145:61–63.

Pinsker HM (1977): Aplysia bursting neurons as endogenous oscil-lators.1. Phase-response curves for pulsed inhibitory synapticinput. J Neurophysiol 40:527–543.

Popovych OV, Tass PA (2010): Synchronization control of interact-ing oscillatory ensembles by mixed nonlinear delayed feed-back. Phys Rev E 82:026204.

Popovych OV, Tass PA (2012): Desynchronizing electrical andsensory coordinated reset neuromodulation. Front Hum Neu-rosci 6:58.

Prinz AA, Thirumalai V, Marder E (2003): The functional conse-quences of changes in the strength and duration of synapticinputs to oscillatory neurons. J Neurosci 23:943–954.

Rauschecker JP (2005): Neural encoding and retrieval of soundsequences. Ann N Y Acad Sci 1060:125–135.

Schenk S, Lamm K, Gundel H, Ladwig KH (2005): [Neurofeed-back-based EEG alpha and EEG beta training. Effectiveness inpatients with chronically decompensated tinnitus]. HNO53:29–37.

Scherg M, Ille N, Bornfleth H, Berg P (2002): Advanced tools fordigital EEG review: Virtual source montages, whole-head map-ping, correlation, and phase analysis. J Clin Neurophysiol19:91–112.

Schlee W, Hartmann T, Langguth B, Weisz N (2009a): Abnormalresting-state cortical coupling in chronic tinnitus. BiomedChromatogr Neurosci 10:11.

Schlee W, Mueller N, Hartmann T, Keil J, Lorenz I, Weisz N (2009b):Mapping cortical hubs in tinnitus. Biomed Chromatogr Biol 7:80.

Schutte NS, Noble W, Malouff JM, Bhullar N (2009): Evaluation ofa model of distress related to tinnitus. Int J Audiol 48:428–432.

Silchenko AN, Adamchic I, Hauptmann C, Tass PA (2013): Impact ofacoustic coordinated reset neuromodulation on effective connectiv-ity in a neural network of phantom sound. Neuroimage 22:133–147.

Smits M, Kovacs S, de Ridder D, Peeters RR, van Hecke P,Sunaert S (2007): Lateralization of functional magnetic reso-nance imaging (fMRI) activation in the auditory pathway ofpatients with lateralized tinnitus. Neuroradiology 49:669–679.

Steriade M (2006): Grouping of brain rhythms in corticothalamicsystems. Neuroscience 137:1087–1106.

Sterman MB, Friar L (1972): Suppression of seizures in an epilep-tic following sensorimotor eeg feedback training. Electroence-phalogr Clin Neurophysiol 33:89–95.

Tass PA (2003): A model of desynchronizing deep brain stimula-tion with a demand-controlled coordinated reset of neural sub-populations. Biol Cybern 89:81–88.

Tass PA, Hauptmann C (2009): Anti-kindling achieved by stimulation tar-geting slow synaptic dynamics. Restor Neurol Neurosci 27:589–609.

Tass PA, Majtanik M (2006): Long-term anti-kindling effects ofdesynchronizing brain stimulation: A theoretical study. BiolCybern 94:58–66.

Tass PA, Popovych OV (2012): Unlearning tinnitus-related cere-bral synchrony with acoustic coordinated reset stimulation:Theoretical concept and modelling. Biol Cybern 106:27–36.

Tass PA, Silchenko A, Hauptmann C, Barnikol U, Speckmann EJ(2009): Long-lasting desynchronization in rat hippocampal sliceinduced by coordinated reset stimulation. Phys Rev E 80: 011902.

Tass PA, Adamchic I, Freund HJ, von Stackelberg T, HauptmannC (2012a): Counteracting tinnitus by acoustic coordinated resetneuromodulation. Restor Neurol Neurosci 30:137–159.

Tass PA, Qin L, Hauptmann C, Dovero S, Bezard E, Boraud T,Meissner WG (2012b): Coordinated reset has sustained after-effects in Parkinsonian monkeys. Ann Neurol 72:816–820.

Thatcher RW (2010): Validity and reliability of quantitative elec-troencephalography. J Neurother 14:122–152.

Turner D, Schunemann HJ, Griffith LE, Beaton DE, Griffiths AM,Critch JN, Guyatt GH (2010): The minimal detectable changecannot reliably replace the minimal important difference. JClin Epidemiol 63:28–36.

van der Loo E, Gais S, Congedo M, Vanneste S, Plazier M,Menovsky T, Van de Heyning P, De Ridder D (2009): Tinnitusintensity dependent gamma oscillations of the contralateral au-ditory cortex. PLoS One 4:e7396.

Vandervoet H (1994): Comparing the predictive accuracy of mod-els using a simple randomization test. Chemometr Intell LabSyst 25:313–323.

Vanneste S, Plazier M, der Loo E, de Heyning PV, Congedo M,De Ridder D (2010): The neural correlates of tinnitus-relateddistress. Neuroimage 52:470–480.

Vanneste S, Plazier M, van der Loo E, Van de Heyning P, De Rid-der D (2011a): The difference between uni- and bilateral audi-tory phantom percept. Clin Neurophysiol 122:578–587.

Vanneste S, van de Heyning P, De Ridder D (2011b): The neuralnetwork of phantom sound changes over time: A comparisonbetween recent-onset and chronic tinnitus patients. Eur J Neu-rosci 34:718–731.

r Reversing Pathological Neural Activity in Tinnitus r

r 2117 r

Verkindt C, Bertrand O, Perrin F, Echallier JF, Pernier J (1995):Tonotopic organization of the human auditory-cortex—N100topography and multiple dipole model analysis. Electroence-phalogr Clin Neurophysiol 96:143–156.

Weisz N, Moratti S, Meinzer M, Dohrmann K, Elbert T (2005):Tinnitus perception and distress is related to abnormal sponta-neous brain activity as measured by magnetoencephalography.PLoS Med 2:e153.

Weisz N, Hartmann T, Dohrmann K, Schlee W, Norena A (2006):High-frequency tinnitus without hearing loss does not meanabsence of deafferentation. Hear Res 222:108–114.

Weisz N, Dohrmann K, Elbert T (2007a): The relevance of sponta-neous activity for the coding of the tinnitus sensation. Tinni-tus: Pathophysiol Treat 166:61–70.

Weisz N, Muller S, Schlee W, Dohrmann K, Hartmann T, Elbert T(2007b): The neural code of auditory phantom perception. JNeurosci 27:1479–1484.

Weisz N, Hartmann T, Muller N, Lorenz I, Obleser J (2011): Alpharhythms in audition: Cognitive and clinical perspectives. FrontPsychol 2:73.

Wold S (1978): Cross-validatory estimation of number of compo-nents in factor and principal components models. Technomet-rics 20:397–405.

Wold S, Eriksson L, Trygg J, Kettaneh N (2004): The PLSmethod—Partial Least Squares projection to latent structures—and its applications in industrial RDP (research, development,and production). In Tech Rep. Umea University.

r Adamchic et al. r

r 2118 r